Four-way wellhead erosion risk evaluation method
By constructing a multi-condition erosion characterization method and a PI-LSTM prediction model, and combining real-time monitoring information for dynamic correction, the problem of difficulty in obtaining the wellhead wall thickness status in real time was solved, realizing real-time and accurate assessment and prediction of wellhead erosion risk, and improving the safety and reliability of wellhead equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHONGQING UNIVERSITY OF SCIENCE AND TECHNOLOGY
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-01
AI Technical Summary
Existing technologies lack the ability to monitor critical erosion areas at the wellhead in real time, making it difficult to obtain wall thickness status in real time. Manual thickness measurement is lagging and lacks accuracy, failing to reflect the development trend of erosion risk in real time, thus affecting the safety and reliability of wellhead equipment.
A multi-condition erosion characterization method was constructed, a "condition-wall thickness response" sample library was established, a PI-LSTM prediction model was used to predict the wall thickness evolution law, and wall thickness monitoring units were deployed at key locations at the wellhead. Dynamic corrections were made in conjunction with real-time monitoring information to form dynamic wall thickness trend prediction and risk assessment.
It enables dynamic analysis of the wall thickness variation trend of key erosion-affected parts of the wellhead and quantitative determination of risk level, significantly improving the foresight and reliability of wellhead erosion risk identification and providing a scientific basis for wellhead safety management.
Smart Images
Figure CN121960156A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas field fracturing safety monitoring and wellhead equipment risk assessment technology, specifically to a method for assessing the risk of erosion at a four-way wellhead. Background Technology
[0002] With the continuous expansion of shale gas development, fracturing technology is evolving towards ultra-large flow rates, high proppant concentrations, and long-term continuous operations. Under these high-energy flow conditions, the proppant-carrying fracturing fluid passes through the turning chamber inside the four-way wellhead at high speed. This results in abrupt changes in flow direction, enhanced shear layers, and strong turbulence zones. The solid-liquid mixture strongly erodes the metal inner wall, causing the wall thickness in critical areas to continuously thin as construction progresses. Local stress concentration and material fatigue problems gradually become prominent, easily leading to safety accidents such as wellhead leakage, four-way perforation, and valve failure, posing a significant threat to the safety of fracturing operations and the integrity of wellhead equipment.
[0003] Currently, the identification of wellhead erosion risk still mainly relies on manual thickness measurement after well shutdown. This method requires personnel to enter the well site during non-operational periods to manually measure limited locations. The measurement cycle is long and cannot reflect real-time changes in wall thickness during fracturing, resulting in significant information lag. At the same time, manual entry near the wellhead under high pressure and high speed conditions poses a high safety risk; due to the limitations of the wellhead structure, the number of measurable points is limited, making it difficult to obtain the spatial distribution and temporal evolution of erosion; in addition, the large temperature fluctuations, strong vibrations, and complex fluid media at the wellhead often increase the error of manual thickness measurement results and have poor repeatability, thereby weakening the reliability of risk assessment.
[0004] Current technologies lack the ability to continuously monitor critical erosion areas at the wellhead and lack methods for predicting wall thickness evolution in real time based on changes in operating conditions. Therefore, they cannot capture the wall thickness change process in real time, nor can they reflect the development trend of erosion. This makes it difficult to identify risk evolution in a timely manner and to quantify the staged evolution and remaining life of erosion risk. Currently, there is a lack of a wellhead erosion risk assessment technology that combines real-time, quantitative, and forward-looking capabilities. This technology should be able to continuously acquire wall thickness information of critical areas under complex fracturing conditions and combine it with changes in operating conditions to determine erosion trends, thereby providing a reliable basis for safe operation and maintenance decisions of wellhead equipment. Summary of the Invention
[0005] This invention provides a method for assessing the risk of wellhead erosion in four-way wells, which solves the problems in the prior art such as the difficulty in obtaining the wellhead wall thickness status in real time, the lag and insufficient accuracy of manual thickness measurement, and the unclear wall thickness evolution trend.
[0006] The technical solution is as follows: A method for assessing the risk of wellhead erosion in a four-way well, the key points of which include the following steps:
[0007] S1: Construct a multi-condition erosion characterization method and establish a "condition-wall thickness response" sample library;
[0008] S2: Based on the "working condition-wall thickness response" sample library in S1, construct a prediction model that reflects the wall thickness evolution law, so that it outputs the wall thickness reduction trend of key corrosion-affected parts under different working conditions.
[0009] S3: Install wall thickness monitoring units at key erosion sites at the wellhead to obtain monitoring information;
[0010] S4: Integrate the predicted remaining wall thickness trend in S2 and the monitoring information in S3 to establish a dynamic correction mechanism, so that the wall thickness change trend can be dynamically updated with the changes in fracturing conditions to obtain the corrected wall thickness trend.
[0011] S5: Based on the wall thickness trend corrected in S4, quantitative assessment of erosion risk and life prediction are carried out in combination with material properties and pressure conditions.
[0012] Preferably, in step S1, key working condition parameters such as displacement, sand ratio, proppant particle size, liquid viscosity and action time are extracted from the fracturing operation record to form a working condition feature set, construct a multi-working condition erosion characterization system, and construct a "working condition-wall thickness response" sample library by combining dismantling and inspection records, experimental tests and simulation results.
[0013] Preferably, the prediction model in step S2 is a PI-LSTM prediction model that integrates the physical mechanism of fluid erosion and deep learning algorithms. This model includes an input layer, a hidden layer, a fully connected layer and an output layer connected in sequence.
[0014] The input layer sets a sliding time window length of T, and it handles multiple consecutive time points t1, t2, ... t3. n Fracturing condition data were collected at "current time t and the past T times" to form a time-series characteristic sequence of operating conditions. ;
[0015] The time-series operating condition characteristic sequence at each moment The data is fed into a hidden layer, which consists of at least two stacked LSTM units, each with 128 neurons. The LSTM units control the flow and retention of erosion information through forget gates, input gates, and output gates to generate the final hidden state at each time step. ;
[0016] Output the final hidden state at each time step of the hidden layer. The data is fed into a ReLU-activated fully connected layer for processing, and then fed into the output layer, which outputs the predicted wall thickness reduction per unit time step at each time step. .
[0017] Preferably: the predicted wall thickness reduction per unit time at each moment is used to calculate the predicted remaining wall thickness at the wellhead, and the formula is as follows:
[0018] ;
[0019] in, To predict the remaining wall thickness at the wellhead, This is the initial wall thickness. Let be the predicted wall thickness reduction per unit time at time t. The time step for data acquisition between two adjacent operating conditions. From time 1 to time t i The actual wall thickness reduction at each moment is summed up.
[0020] The predicted remaining wall thickness at each time point is sequentially connected along the time dimension to form the trend of the predicted remaining wall thickness change.
[0021] Preferably: the time-series operating condition feature sequence The operating conditions at each moment are as follows:
[0022]
[0023] in, The fracturing fluid discharge rate (m³) at time t 3 / min); The proppant sand ratio (%) at time t; The average particle size of the proppant is (mm). Fracturing fluid viscosity (mPa.s); The instantaneous kinetic energy factor is based on the erosion physics formula. Derivative, the calculation formula is as follows This is used to explicitly guide neural networks to focus on high-energy erosion regions; This represents the cumulative amount of sand added up to time t.
[0024] Preferably: the memory and update of timing information in the hidden layer are accomplished through the forward propagation computation logic of the LSTM unit, and the forward propagation computation logic is as follows:
[0025] (Forgotten Gate)
[0026] (Input Gate)
[0027] (Candidate status)
[0028] (Cell status update)
[0029] (Output Gate)
[0030] (Hidden state)
[0031] in It is the Sigmoid activation function. For Hadamard product, W and b are the weight matrix and bias term obtained during training.
[0032] Preferably, in step S3, key erosion areas are selected at locations of abrupt changes in flow direction, sharp turns, and concentrated particle impact within the wellhead, and wall thickness monitoring units are installed on their outer walls. These monitoring units record the real-time wall thickness values at various times. It is continuously transmitted to the ground monitoring terminal.
[0033] Preferably, the correction mechanism in step S4 uses a "sliding window error compensation method" for correction, specifically:
[0034] Instantaneous residual The monitoring terminal will receive the measured values uploaded by the wall thickness monitoring unit in real time at each moment. The predicted remaining wall thickness at the wellhead at the corresponding time. Subtraction yields the instantaneous residual Its formula is:
[0035] ;
[0036] Filtering and Trend Extraction: The residual series at each time point is processed by exponentially weighted moving average (EWMA) to obtain the smoothed error trend. Its formula is:
[0037] ;
[0038] in, This is a smoothing factor, with a value range of 0.1 to 0.3. Let be the instantaneous residual at time t. The smoothing error trend accumulated over time;
[0039] Dynamic fusion correction: smoothing the error trend at each time step. This feedback is used to input the corresponding prediction results to obtain the final corrected wall thickness. ;
[0040] ;
[0041] in, This represents the trust weighting coefficient, which increases the trust level when the monitored data is continuous and the fluctuations are within a preset, physically reasonable range. Value; when the monitored data shows abnormal fluctuations, the system automatically reduces it. This value allows the output to smoothly transition to the model's predicted value;
[0042] The corrected wall thickness at each time point By sequentially connecting them along the time dimension, the trend of wall thickness variation after correction is formed.
[0043] Preferably: if a smoothing error trend is detected If the load exceeds a preset threshold for N consecutive minutes, online incremental learning of the model is triggered. The working condition data of the most recent M minutes and the measured wall thickness are stored in a buffer, and the weights W of the fully connected layer of the LSTM network are adjusted using the backpropagation algorithm. out Fine-tuning and updating are performed to allow the model to proactively adapt to the current erosion characteristics.
[0044] Preferably, step S5 is based on the final corrected wall thickness variation trend described above, and combines fluid mechanics and materials mechanics criteria to perform multi-dimensional risk quantification.
[0045] Calculate the maximum safe working pressure for the current wall thickness using the Barlow formula. Its formula is:
[0046] ;
[0047] in, The yield strength (MPa) of the wellhead crossbeam material. To design a safety factor, The corrected wall thickness. The outer diameter of the corroded part of the four-way connector ( );
[0048] Risk coefficient Based on the maximum safe working pressure under the current wall thickness and real-time collected pump pressure The formula is calculated as follows:
[0049]
[0050] Among them, the safe status (green): <0.8 and > ;
[0051] Warning status (yellow): 0.8≤ ≤0.95;
[0052] Dangerous state (red) 0.95≤ or ≤ ;
[0053] Remaining life (RUL) prediction is the prediction of wall thickness reduction to the minimum allowable limit. Time required Its formula is:
[0054]
[0055] in, You can take the average erosion rate over the past W minutes, or the rate predicted by the PI-LSTM prediction model for future operating conditions.
[0056] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a multi-condition erosion characterization method, a wall thickness prediction model, a real-time monitoring and correction mechanism, and a risk quantitative evaluation system, this invention achieves dynamic analysis of the wall thickness variation trend and quantitative determination of the risk level of key eroded parts at the four-way wellhead. Compared with the traditional method that relies on manual thickness measurement, this invention can reflect the wall thickness reduction status more timely, comprehensively, and accurately, significantly improving the foresight and reliability of wellhead erosion risk identification during fracturing, and providing efficient and scientific technical support for wellhead safety management. Attached Figure Description
[0057] Figure 1 A flowchart illustrating the steps of a four-way wellhead erosion risk assessment method;
[0058] Figure 2 A schematic diagram of the key eroded areas and wall thickness monitoring system at the four-way wellhead;
[0059] Figure 3 A schematic diagram showing the volume of fluids and proppant during the fracturing pumping stage;
[0060] Figure 4 A schematic diagram showing the relationship between proppant particle size, erosion rate and erosion depth at different stages;
[0061] Figure 5 This is a schematic diagram showing the evolution of wall thickness and erosion risk zoning of key erosion areas at the four-way wellhead. Detailed Implementation
[0062] The present invention will be further described below with reference to the embodiments and accompanying drawings.
[0063] like Figure 1 As shown, a method for assessing the risk of wellhead erosion in a four-way well includes the following steps:
[0064] S1: Construct a multi-condition erosion characterization method and establish a "condition-wall thickness response" sample library;
[0065] S2: Based on the "working condition-wall thickness response" sample library in S1, construct a prediction model that reflects the wall thickness evolution law, so that it outputs the wall thickness reduction trend of key corrosion-affected parts under different working conditions.
[0066] S3: Install wall thickness monitoring units at key erosion sites at the wellhead to obtain monitoring information;
[0067] S4: Integrate the predicted remaining wall thickness trend in S2 and the monitoring information in S3 to establish a dynamic correction mechanism, so that the wall thickness change trend can be dynamically updated with the changes in fracturing conditions to obtain the corrected wall thickness trend.
[0068] S5: Based on the wall thickness trend corrected in S4, quantitative assessment of erosion risk and life prediction are carried out in combination with material properties and pressure conditions.
[0069] This invention aims to assess the erosion risk of fracturing wellheads. It divides the entire analysis process into five stages: operational condition modeling, wall thickness prediction, key component monitoring, model correction, and risk assessment. This forms a complete process chain of "modeling—prediction—monitoring—correction—assessment," enabling dynamic analysis of wall thickness variation trends and quantitative determination of risk levels in key erosion-affected areas of the wellhead. It can more timely, comprehensively, and accurately reflect the wall thickness reduction status, significantly improving the foresight and reliability of wellhead erosion risk identification during fracturing, and providing efficient and scientific technical support for wellhead safety management.
[0070] S1: Construct a multi-condition erosion characterization method and establish a "condition-wall thickness response" sample library;
[0071] In step S1, key operating condition parameters such as displacement, sand ratio, proppant particle size, liquid viscosity and action time are extracted from the fracturing operation records, and a "operating condition-wall thickness response" sample library is constructed by combining dismantling and inspection records, experimental tests and simulation results.
[0072] Specifically, by reading original data such as on-site discharge volume, sand ratio, proppant particle size, liquid viscosity, and duration of each stage, and dividing the construction into typical working condition sections such as pre-filling liquid, sand addition, incremental sand ratio, constant sand ratio, and liquid replacement, a set of working condition characteristics that can describe the changes in solid-liquid kinetic energy levels is formed, a multi-working condition erosion characterization system is constructed, and a "working condition-wall thickness response" sample library is constructed by combining the results of dismantling and testing thickness, indoor erosion experiments, or numerical simulation results such as CFD.
[0073] Figure 3 This is a schematic diagram of the liquid and proppant volume ledger during the construction phase, showing the changes in injection volume and sand ratio over time, thus reflecting the differences in erosion intensity at different stages. Combined with... Figure 4 The relationship between proppant particle size, erosion rate and erosion depth shown can be further extracted from the data to extract erosion behavior characteristics under different particle size combinations, making multi-condition characterization trainable and distinguishable.
[0074] S2: Based on the "working condition-wall thickness response" sample library in S1, construct a prediction model that reflects the wall thickness evolution law, so that it outputs the wall thickness reduction trend of key corrosion-affected parts under different working conditions.
[0075] Step S2 does not employ traditional linear regression or a single physical formula. Instead, it establishes a Physics-Informed LSTM (PI-LSTM) prediction model that integrates the physical mechanism of fluid erosion with deep learning algorithms. Since fracturing erosion has a significant time-cumulative effect—that is, the wall thickness damage at the current moment is related to the cumulative operating conditions over a past period—a Long Short-Term Memory (LSTM) network is chosen to capture this temporal dependency.
[0076] The PI-LSTM prediction model comprises an input layer, a hidden layer, a fully connected layer, and an output layer connected in sequence.
[0077] The input layer sets a sliding time window length of T, and it handles multiple consecutive time points t1, t2, ... t3. n Fracturing condition data were collected at "current time t and the past T times" to form a time-series characteristic sequence of operating conditions. ;
[0078] The time-series operating condition feature sequence The operating conditions at each moment are as follows:
[0079]
[0080] in, The fracturing fluid discharge rate (m³) at time t 3 / min); The proppant sand ratio (%) at time t; The average particle size of the proppant is (mm). Fracturing fluid viscosity (mPa.s); The instantaneous kinetic energy factor is based on the erosion physics formula. Derivative, the calculation formula is as follows This is used to explicitly guide neural networks to focus on high-energy erosion regions; This represents the cumulative amount of sand added up to time t.
[0081] The time-series operating condition characteristic sequence at each moment The information is fed into a hidden layer, which consists of at least two stacked LSTM units, each with 128 neurons. This multi-layered structure enhances the ability to capture complex temporal relationships. The LSTM units control the flow of information through forget gates, input gates, and output gates to generate the final hidden state at each time step. ;
[0082] The hidden layer uses the forward propagation logic of LSTM units to complete the memorization and updating of temporal information. Forward propagation is a time-progression process, and the hidden state at each time t... It depends on the hidden state of the previous moment t-1. The specific calculation logic, in conjunction with the current operating condition feature sequence as input, is as follows:
[0083] (Forgotten Gate)
[0084] (Input Gate)
[0085] (Candidate status)
[0086] (Cell status update)
[0087] (Output Gate)
[0088] (Hidden state)
[0089] in It is the Sigmoid activation function. For Hadamard product, W and b are the weight matrix and bias term obtained during training.
[0090] When there is no historical record at the initial time (i.e.) In the scenario of "hidden state initialization", the recursion is started by "hidden state initialization", and the hidden state is initialized as an all-zero vector as the starting point.
[0091] The forget gate filters the erosion time-series information from historical moments, removing invalid historical erosion information that has no impact on the current wall thickness reduction, and retaining valid historical erosion memories. The input gate filters the current operating condition characteristics, including only valid current erosion information that significantly affects wall thickness reduction in the calculation. The candidate cell state quantifies the intensity and properties of the current valid information, clarifying the aggravating or mitigating effect of the current operating conditions on erosion. The cell state update merges the retained valid historical erosion memories with the quantified valid current erosion information to form a core memory bank containing the cumulative effect of erosion throughout the entire time series. The output gate performs a secondary filtering on the erosion information in the core memory bank, extracting core erosion features that can be directly used for wall thickness reduction calculation, and converting them into hidden states. .
[0092] Output the hidden state at each time step of the hidden layer. The hidden state is passed into the ReLU-activated fully connected layer. Perform a nonlinear transformation to extract key features, process them, and then feed them into the output layer to output the predicted wall thickness reduction per unit time at each time step. .
[0093] The predicted wall thickness reduction per unit time at each moment is used to calculate the predicted remaining wall thickness at the wellhead. The formula is as follows:
[0094] ;
[0095] in, To predict the remaining wall thickness at the wellhead, This is the initial wall thickness. Let be the predicted wall thickness reduction per unit time at time t. The time step for data acquisition between two adjacent operating conditions. From time 1 to time t i The actual wall thickness reduction at each moment is summed up.
[0096] By sequentially connecting the remaining wall thickness at each time point along the time dimension, a trend of remaining wall thickness change is formed, providing a priori predictive capability for risk analysis.
[0097] S3: Install wall thickness monitoring units at key erosion sites at the wellhead to obtain monitoring information;
[0098] like Figure 2 As shown, key erosion areas are selected at locations of abrupt changes in flow direction, sharp turns, and concentrated particle impact within the wellhead, and wall thickness monitoring units are deployed on their outer walls. Monitoring methods can be selected based on site adaptability, such as ultrasonic thickness measurement, radio frequency thickness measurement, or surface acoustic wave sensing technology. The monitoring units will record the real-time wall thickness values at various times. The data is continuously transmitted to the ground monitoring terminal, forming a dynamic acquisition mechanism for the in-service wall thickness, providing real-time input for the model correction process.
[0099] S4: Integrate the predicted remaining wall thickness trend in S2 and the monitoring information in S3 to establish a dynamic correction mechanism, so that the wall thickness change trend can be dynamically updated with the changes in fracturing conditions to obtain the corrected wall thickness trend.
[0100] Although the PI-LSTM prediction model in S2 was trained on historical data, it may still produce deviations when faced with unknown extreme working conditions or sensor drift. Therefore, in S4, real-time monitoring data is introduced as a "supervisory signal" to construct a closed-loop feedback correction mechanism. This correction mechanism uses the "sliding window error compensation method" for correction, specifically:
[0101] Instantaneous residual The monitoring terminal will receive the measured values uploaded by the wall thickness monitoring unit in real time at each moment. The predicted remaining wall thickness at the wellhead at the corresponding time. Subtraction yields the instantaneous residual Its formula is:
[0102] ;
[0103] Filtering and Trend Extraction: The residual series at each time point is processed by exponentially weighted moving average (EWMA) to obtain the smoothed error trend. Its formula is:
[0104] ;
[0105] in, This is a smoothing factor, with a value range of 0.1 to 0.3. Let be the instantaneous residual at time t. For the historical accumulation of smoothed error trends, when calculating the first smoothed error trend If there is no historical accumulation of smoothing error trend, the default value is 0;
[0106] Dynamic fusion correction: smoothing the error trend at each time step. This feedback is used to input the corresponding prediction results to obtain the final corrected wall thickness. ;
[0107] ;
[0108] in, This represents the trust weighting coefficient, which increases the trust level when the monitored data is continuous and the fluctuations are within a preset, physically reasonable range. The value is within the reasonable range of 0-1;
[0109] When the monitoring data shows abnormal fluctuations, the system automatically reduces... The value is used to smoothly transition the output to the model prediction value and prevent false alarms. Its abnormal jump refers to the spike generated by the sensor due to electromagnetic interference, that is, the monitoring data shows a step that exceeds the theoretical limit or data loss in adjacent time steps.
[0110] The corrected wall thickness at each time point By sequentially connecting them along the time dimension, the trend of wall thickness variation after correction is formed.
[0111] If a smoothing error trend is detected If the load exceeds a preset threshold for N consecutive minutes, online incremental learning of the model is triggered. The working condition data of the most recent M minutes and the measured wall thickness are stored in a buffer, and the weights W of the fully connected layer of the LSTM network are adjusted using the backpropagation algorithm. out Fine-tuning and updating are performed to allow the model to proactively adapt to the current erosion characteristics.
[0112] The correction mechanism can be based on online correction of model residuals, or it can achieve real-time fitting between the predicted curve and the measured curve through data fusion methods such as Kalman filtering, so that the prediction results can be adjusted in a timely manner according to changes in operating conditions. For example, when the sand ratio increases or the displacement changes abruptly, causing the wall thickness to decrease at a faster rate, the monitoring data will immediately prompt the prediction curve to be corrected accordingly, ensuring that the model always reflects the actual reduction trend. Figure 5 The diagram shows the wall thickness evolution trend after the prediction-measurement fusion, and it can be seen that the corrected curve can closely match the actual change trajectory.
[0113] S5: Based on the wall thickness trend modified in S4, quantitative assessment of erosion risk and life prediction are carried out in combination with material properties and pressure conditions.
[0114] Calculate the maximum safe working pressure for the current wall thickness using the Barlow formula. Its formula is:
[0115] ;
[0116] in, The yield strength (MPa) of the wellhead crossbeam material. To design a safety factor, The corrected wall thickness. The outer diameter of the corroded part of the four-way connector ( );
[0117] Risk coefficient Based on the maximum safe working pressure under the current wall thickness and real-time collected pump pressure The formula is calculated as follows:
[0118]
[0119] Among them, the safe status (green): <0.8 and > ;
[0120] Warning status (yellow): 0.8≤ ≤0.95;
[0121] Dangerous state (red) 0.95≤ or ≤ ;
[0122] Remaining lifetime (RUL) prediction is based on a revised... Starting from the current point, the PI-LSTM prediction model, calibrated with real-time data, is used to predict future operating conditions and extrapolate future wall thickness trends and remaining lifespan, ultimately predicting when the wall thickness will decrease to the minimum permissible limit. Time required Its formula is:
[0123]
[0124] in, The average erosion rate over the past W minutes can be used. The past W minutes can be understood as a continuous time period of W minutes backwards from the current time t, where W is a time length parameter customized based on the fluctuation frequency of the fracturing operation conditions. The erosion rate is obtained by calculating the average of the corrected wall thickness reduction per unit time at all moments within this time period. To avoid the interference of transient fluctuations on the remaining lifetime prediction results;
[0125] Alternatively, the predicted rate of future working conditions can be obtained from the PI-LSTM prediction model. The temporal variation law of fracturing working conditions learned by the PI-LSTM prediction model can be used to input the current and historical working condition sequences to predict the working condition parameters for a period of time in the future. Then, the future working condition parameters can be input into the LSTM wall thickness reduction model to obtain the wall thickness reduction per unit time at each future moment, and thus obtain the prediction rate. This can make the remaining life prediction result fit the actual erosion intensity of subsequent construction.
[0126] Both the average erosion rate over the past W minutes and the predicted rate for future operating conditions are based on the calibrated LSTM model and the current instantaneous erosion rate as the core benchmark. (The instantaneous erosion rate is obtained by modifying the original output of the PI-LSTM prediction model through a correction mechanism) The average erosion rate over the past W minutes is smoothed by the "recent average" to ensure the reliability of predictions under stable operating conditions; future operating condition predictions are aligned with trends by the "future forecast" to ensure the accuracy of predictions under fluctuating operating conditions. The two can be flexibly switched according to the on-site operating conditions, ensuring both the reliability of predictions under stable operating conditions and the accuracy of predictions under fluctuating operating conditions, thus achieving broad adaptability to operating conditions and high prediction accuracy.
[0127] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. Those skilled in the art, under the guidance of the present invention, can make various similar representations without departing from the spirit and claims of the present invention, and such modifications all fall within the protection scope of the present invention.
Claims
1. A method for assessing the risk of wellhead erosion in a four-way well, characterized in that: Includes the following steps: S1: Construct a multi-condition erosion characterization method and establish a "condition-wall thickness response" sample library; S2: Based on the "working condition-wall thickness response" sample library in S1, construct a prediction model that reflects the wall thickness evolution law, so that it outputs the wall thickness reduction trend of key corrosion-affected parts under different working conditions. S3: Install wall thickness monitoring units at key erosion sites at the wellhead to obtain monitoring information; S4: Integrate the predicted remaining wall thickness trend in S2 and the monitoring information in S3 to establish a dynamic correction mechanism, so that the wall thickness change trend can be dynamically updated with the changes in fracturing conditions to obtain the corrected wall thickness trend. S5: Based on the wall thickness trend corrected in S4, quantitative assessment of erosion risk and life prediction are carried out in combination with material properties and pressure conditions.
2. The method for assessing the risk of wellhead erosion according to claim 1, characterized in that: In step S1, key operating condition parameters such as displacement, sand ratio, proppant particle size, liquid viscosity and action time are extracted from the fracturing operation records to form an operating condition feature set, construct a multi-operating condition erosion characterization system, and build a "operating condition-wall thickness response" sample library by combining dismantling and inspection records, experimental tests and simulation results.
3. The method for assessing the risk of wellhead erosion according to claim 2, characterized in that: The prediction model in step S2 is a PI-LSTM prediction model that integrates the physical mechanism of fluid erosion and deep learning algorithms. This model includes an input layer, a hidden layer, a fully connected layer and an output layer connected in sequence. The input layer sets a sliding time window length of T, and it handles multiple consecutive time points t1, t2, ... t3. n Fracturing condition data for "the current time t and the past T times" are collected separately to form a time-series characteristic sequence of operating conditions. ; The time-series operating condition characteristic sequence at each moment The data is fed into a hidden layer, which consists of at least two stacked LSTM units, each with 128 neurons. The LSTM units control the flow and retention of erosion information through forget gates, input gates, and output gates to generate the final hidden state at each time step. ; Output the final hidden state at each time step of the hidden layer. The data is fed into a ReLU-activated fully connected layer for processing, and then fed into the output layer, which outputs the predicted wall thickness reduction per unit time step at each time step. .
4. The method for assessing the risk of wellhead erosion according to claim 3, characterized in that: The predicted wall thickness reduction per unit time at each moment is used to calculate the predicted remaining wall thickness at the wellhead. The formula is as follows: ； in, To predict the remaining wall thickness at the wellhead, This is the initial wall thickness. Let be the predicted wall thickness reduction per unit time at time t. The time step for data acquisition between two adjacent operating conditions. From time 1 to time t i The actual wall thickness reduction at each moment is summed up. The predicted remaining wall thickness at each time point is sequentially connected along the time dimension to form the trend of the predicted remaining wall thickness change.
5. The method for assessing the risk of wellhead erosion according to claim 3, characterized in that: The time-series operating condition feature sequence The operating conditions at each moment are as follows: in, The fracturing fluid discharge rate (m³) at time t 3 / min); The proppant sand ratio (%) at time t; The average particle size of the proppant is (mm). Fracturing fluid viscosity (mPa.s); The instantaneous kinetic energy factor is based on the erosion physics formula. Derivative, the calculation formula is as follows This is used to explicitly guide neural networks to focus on high-energy erosion regions; This represents the cumulative amount of sand added up to time t.
6. A method for assessing the risk of wellhead erosion according to claim 3 or 5, characterized in that: The hidden layer uses the forward propagation computation logic of the LSTM unit to complete the memorization and updating of timing information. The forward propagation computation logic is as follows: (Forgotten Gate) (Input Gate) (Candidate status) (Cell status update) (Output Gate) (Hidden state) in It is the Sigmoid activation function. For Hadamard product, W and b are the weight matrix and bias term obtained during training.
7. The method for assessing the risk of wellhead erosion according to claim 3, characterized in that: In step S3, key erosion areas are selected at locations of abrupt changes in flow direction, sharp turns, and concentrated particle impact within the wellhead. Wall thickness monitoring units are then installed on the outer wall of these areas. These monitoring units record the real-time wall thickness values at various times. It is continuously transmitted to the ground monitoring terminal.
8. The method for assessing the risk of wellhead erosion according to claim 7, characterized in that: The correction mechanism in step S4 uses a "sliding window error compensation method" for correction, specifically as follows: Instantaneous residual The monitoring terminal will receive the measured values uploaded by the wall thickness monitoring unit in real time at each moment. The predicted remaining wall thickness at the wellhead at the corresponding time. Subtraction yields the instantaneous residual Its formula is: ； Filtering and Trend Extraction: The residual series at each time point is processed by exponentially weighted moving average (EWMA) to obtain the smoothed error trend. Its formula is: ； in, This is a smoothing factor, with a value range of 0.1 to 0.
3. Let be the instantaneous residual at time t. The smoothing error trend accumulated over time; Dynamic fusion correction: smoothing the error trend at each time step. This feedback is used to input the corresponding prediction results to obtain the final corrected wall thickness. ; ； in, This represents the trust weighting coefficient, which increases the trust level when the monitored data is continuous and the fluctuations are within a preset, physically reasonable range. Value; when the monitored data shows abnormal fluctuations, the system automatically reduces it. This value allows the output to smoothly transition to the model's predicted value; The corrected wall thickness at each time point By sequentially connecting them along the time dimension, the trend of wall thickness variation after correction is formed.
9. The method for assessing the risk of wellhead erosion according to claim 8, characterized in that: If a smoothing error trend is detected If the preset threshold is exceeded for N consecutive minutes, online incremental learning of the model is triggered. The working condition data of the most recent M minutes and the measured wall thickness are stored in a buffer. The weights W of the fully connected layer of the LSTM network are then adjusted using the backpropagation algorithm, which is the opposite of forward propagation. out Fine-tuning and updating are performed to allow the model to proactively adapt to the current erosion characteristics.
10. The method for assessing the risk of wellhead erosion according to claim 9, characterized in that: Step S5 is based on the final corrected wall thickness change trend mentioned above, and combines fluid mechanics and material mechanics criteria to perform multi-dimensional risk quantification. Calculate the maximum safe working pressure for the current wall thickness using the Barlow formula. Its formula is: ； in, The yield strength (MPa) of the wellhead crossbeam material. To design a safety factor, The corrected wall thickness. The outer diameter of the corroded part of the four-way connector ( ); Risk coefficient Based on the maximum safe working pressure under the current wall thickness and real-time collected pump pressure The formula is calculated as follows: Among them, the safe status (green): <0.8 and > ; Warning status (yellow): 0.8≤ ≤0.95; Dangerous state (red) 0.95≤ or ≤ ; Remaining useful life (RUL) prediction is based on the prediction that the wall thickness will decrease to the minimum allowable limit. Time required Its formula is: in, You can take the average erosion rate over the past W minutes, or the rate predicted by the PI-LSTM prediction model for future operating conditions.